Thursday, May 7, 2009

NVidia's Tesla C1060 - The FM principle?

I've been hearing a lot about NVidia's new "super computer on your desk", but all I can really find is marketing talk. I was skeptical of the claims until I went to the site and looked at the specs. They really do look impressive. 240 cores, each running at almost 2ghz with 4GB of dedicated memory. That's a lot of hardware, especially when you see some "super computers" on sale that have four of them in one machine. 

There's just one thing that I can't figure out...how does it work? According to what little documentation I can find, the system processor offloads work onto the GPUs, but I'm a little fuzzy on how that works. Since there are drivers, maybe they offload intensive work onto the card? I'm not sure, honestly. 

Does anyone have any experience with this? We do some large math here that might be handy to offload to a card if it works well enough. A price tag of $1300 is a bit much for me to experiment with at the moment, though.